Supplement to A Framework for Making Face Detection Benchmark Databases
The following links offer you the datasets used in the experiments reported in the paper. The first one contains two subsets, one is the MIT/CMU dataset with faces replaced by those cropped from CMU PIE, and the other is the FDDB dataset with faces replaced in the same way.
The second one contains 5 subsets, namely Pose, Illum_Front, Illum_Down, Occlusion and Orientation. Each subset contained images with complex backgrounds and faces with specific parameters.
For details, please refer the paper.
Link : Datasets_with_Faces_Replaced